The paper "OntoNotes: The 90% Solution" by Eduard Hovy describes the methodology and results of the OntoNotes project, which aims to create a large, multilingual, richly-annotated corpus with 90% inter-annotator agreement. The project focuses on a domain-independent representation of literal meaning, including predicate structure, word sense, ontology linking, and coreference. The initial release will include 300K words of English and 250K words of Chinese newswire. The methodology involves treebanking, propbanking, word sense disambiguation, ontology linking, and coreference annotation. The treebanking process uses a parsing system to recover predicate-argument structures, while propbanking focuses on verb argument structures. Word sense disambiguation groups fine-grained WordNet senses into coarser-grained senses to improve accuracy and productivity. The ontology links these senses to the Omega ontology, and coreference annotation connects coreferring instances of specific referring expressions. The project extends existing resources like FrameNet and Salsa, aiming to enable automated semantic analysis and providing new training data for related tasks.The paper "OntoNotes: The 90% Solution" by Eduard Hovy describes the methodology and results of the OntoNotes project, which aims to create a large, multilingual, richly-annotated corpus with 90% inter-annotator agreement. The project focuses on a domain-independent representation of literal meaning, including predicate structure, word sense, ontology linking, and coreference. The initial release will include 300K words of English and 250K words of Chinese newswire. The methodology involves treebanking, propbanking, word sense disambiguation, ontology linking, and coreference annotation. The treebanking process uses a parsing system to recover predicate-argument structures, while propbanking focuses on verb argument structures. Word sense disambiguation groups fine-grained WordNet senses into coarser-grained senses to improve accuracy and productivity. The ontology links these senses to the Omega ontology, and coreference annotation connects coreferring instances of specific referring expressions. The project extends existing resources like FrameNet and Salsa, aiming to enable automated semantic analysis and providing new training data for related tasks.